Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
65 lines (44 loc) · 2.16 KB

CONTRIBUTING.md

File metadata and controls

65 lines (44 loc) · 2.16 KB
Raw

Contributing to TetriServe

Thanks for your interest in contributing! Here's everything you need to get started.

Project Structure

  • tetriserve/ — core library (scheduler, runner, server, globals)
  • benchmark/ — SLO attainment benchmark suite
  • tests/ — unit and integration tests
  • doc/ — architecture and API documentation

See doc/architecture.md for a full system overview.

Development Setup

Requirements: Python 3.10+, CUDA 12.4+, 2+ NVIDIA GPUs

git clone https://github.com/DiT-Serving/TetriServe && cd TetriServe

uv venv .venv --python 3.10
uv pip install -p .venv torch==2.6.0 --index-url https://download.pytorch.org/whl/cu124
uv pip install -p .venv -e ".[diffusers]"
uv pip install -p .venv --no-build-isolation "flash-attn==2.6.3"

# Verify
python -c "import tetriserve; print('import OK')"

Code Style

  • Python 3.10+, formatted with black (line length 88)
  • Type hints encouraged for public APIs
  • Pre-commit hooks configured — run pre-commit install after cloning

Key Conventions

  • Multi-GPU: code uses torch.distributed with NCCL; GPU group setup is in tetriserve/globals.py
  • Scheduler: scheduling logic lives in tetriserve/scheduler/; the primary scheduler is dyn_slo_schedule_global_dp
  • Latent transfer: P2P latent handoff between GPU groups is in tetriserve/runner.py; skip_latent_transfer=True uses dummy tensors (valid for scheduler benchmarking)
  • Benchmark configs: YAML files under benchmark/serving_benchmark/configs/ — one baseline and one TetriServe config per hardware target

Testing

# Unit tests (no GPU required)
python -m pytest tests/unit/ -q

# Full tests (requires GPU)
CUDA_VISIBLE_DEVICES=0,1 python -m pytest tests/ -q

Run tests before submitting a PR.

Commit Style

  • One logical change per commit
  • Format: component: brief description (e.g. scheduler: fix SLO window boundary condition)

Submitting a PR

  1. Fork the repo and create a branch from main
  2. Make your changes with tests where applicable
  3. Ensure python -c "import tetriserve" and pytest pass
  4. Open a PR with a clear description of what changed and why